Multi-omics analyses identify mannose phosphate isomerase-centered hypoxia-induced angiogenesis signature in colorectal cancer

Abstract

Background

Hypoxia and angiogenesis are crucial hallmarks of cancer that play key roles in the development and progression of colorectal cancer (CRC). However, the transcriptional mechanism underlying hypoxia induced angiogenesis (HIA) remain elusive. This study aimed to explore the regulatory networks, molecular mechanisms, and prognostic value of HIA-related genes.

Methods

We collected multi-omics data, including chromatin immunoprecipitation sequencing (ChIP-seq), bulk RNA-seq, single cell RNA-seq, spatial transcriptomics, and microarray data from CRC patients and cell lines. Computational methods, including single sample gene set enrichment analysis (ssGSEA), signature-related gene analysis (SRGA), consensus clustering, and others, were utilized to explore the correlation between hypoxia and angiogenesis, identify the HIA-related genes, and establish the risk scoring system based on HIA-related genes. The role of mannose phosphate isomerase (MPI) was validated using quantitative real-time PCR (RT-qPCR), Co-immunoprecipitation (Co-IP), western blot, colony formation, tube formation assay, and subcutaneous xenograft tumor models in vitro and in vivo.

Results

We identified 12 HIA-related genes that are transcriptionally activated by hypoxia-inducible factors (HIFs) and functionally implicated in angiogenesis in CRC. Based on the differentially expressed genes among HIA-related CRC subtypes, we constructed a prognostic scoring system termed HIAscore. Patients with high HIAscore was correlated with poor survival, aggressive phenotype, and immunosuppressive tumor microenvironment. Spatial analysis revealed sequestration regions between epithelial cells with higher HIAscore and T/I/NK cells, hindering their infiltration. Particularly, MPI was found to interact with lactate dehydrogenase A (LDHA), and promote proliferation and angiogenesis of CRC through phosphorylation and activation of Janus kinase 2/signal transducers and activators of transcription 3 (JAK2/STAT3) signaling pathways.

Conclusions

This study depicts the transcriptional landscape linking hypoxia and angiogenesis in CRC, and identifies MPI as a novel regulator of this process.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07291-8.

Background

Colorectal cancer (CRC) is the third predominant cancer and the second leading cause of cancer-related death worldwide. According to the latest global statistics, over 1.9 million new CRC cases and nearly 1 million deaths were reported in 2022, posing a severe threat to the healthcare system [1]. Advances in surgical resection, neoadjuvant chemotherapy, targeted therapy, and immunotherapy have improved the prognosis of CRC patients. However, therapeutic strategies for CRC patients remain contentious, as none of them has achieved consistent clinical efficacy. About 20–45% of these patients undergo tumor recurrence or metastasis, and the 5-year overall survival rate of advanced CRC is still poor, estimated at less than 20% [2]. One possible reason is the lack of understanding to the molecular mechanism behind high heterogeneity among patients, hindering tailored treatment. While previous studies have divided CRC into 4 consensus molecular subtypes (CMS) or 3 pathway-derived subtypes [3, 4], it remains unclear whether additional subtypes can be identified based on specific cancer hallmarks and whether these subgroups would benefited from different treatment strategies. Therefore, it is of great significance to elucidate CRC subtypes and explore therapeutic targets to improve precision medicine.

The rapid proliferation of cancer cells requires adequate oxygen and nutrients. Hypoxia occurs in the core of most solid tumors due to insufficient vascular oxygen supply, leading to a broad metabolic rewiring and tumor microenvironment (TME) reprogramming [5]. Accumulating evidence have suggested that hypoxia is a key driver for cancer development and progression [6]. Elevated expression of hypoxia-inducible factors (HIFs) is a key feature of this phenotype, which orchestrate transcriptional response to low oxygen conditions. HIFs are central transcription factors that bind to the hypoxia-response element in target gene promoters, thereby regulating key genes involved in angiogenesis, such as vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF), stromal cell-derived factor-1 (SDF1, or CXCL12), placental growth factor (PGF), fibroblast growth factor (FGF), and angiopoietin 2 (ANGPT2) [7]. Subsequently, these downstream effectors collectively facilitate endothelial cell sprouting and neovascularization, a process termed angiogenesis. Angiogenesis is a well-recognized hallmark of cancer and play pivotal role in tumor growth, invasion, metastasis, and drug resistance [7]. Multiple signaling pathways have been implicated in cascade hypoxia to angiogenic responses, among which sustained activation of IL6/JAK/STAT3 axis is of particular notable [8–10]. Hyperactivation of JAK2/STAT3 induces expression of VEGF and specific matrix metalloproteinase (MMP), thereby promoting cellular proliferation and invasion, and is regarded as a promising anti-cancer target.

The complicated crosstalk between cancer and non-cancerous cells within the TME plays essential role in development and progression of cancer. Numerous studies have illustrated that hypoxia can impair the innate and adaptative anti-tumor immunity by multiple mechanisms to mediate immune evasion and resistance to treatment [11–13]. Under hypoxic condition, the expression of immune checkpoint molecule PD-L1 increases in a HIF1A-dependent manner across various cell types, including myeloid-derived suppressor cells (MDSCs), tumor-associated macrophages, dendritic cells, and tumor cells. This resulted in deactivation and functional suppression of tumor infiltrated cytotoxic T lymphocytes [14, 15]. In addition, TME undergoes reprogramming during hypoxia driven neovascularization, leading to an immunosuppressive microenvironment. Abnormal tumor vasculature impedes the infiltration of immune cells by inhibiting dendritic cell maturation and antigen presentation, recruiting immunosuppressive cells such as cancer-associated fibroblasts (CAFs), inducing inflammation, and reducing T cell-mediated cytotoxicity [16–18]. Therefore, targeting angiogenic factors to convert immune suppressive tumors into immune activating ones offers a promising avenue for anti-cancer therapy [19, 20].

Despite these advances, our knowledges towards the overall biological roles and the transcriptional mechanisms underlying hypoxia induced angiogenesis (HIA) is still limited. Whether HIA-related genes could be exploited as prognostic and therapeutic targets for CRC remains elusive. In this study, we extensively investigate and explore the importance of HIA-related genes in CRC through comprehensive analyses of multiple datasets and experimental validation. Utilizing chromatin immunoprecipitation (ChIP) sequencing, transcriptome, and several computational methods, we identified 12 genes as putative HIA-related genes. Based on the expression of HIA-related genes, CRC patients were classified into 3 subtypes with distinct tumor hallmark activations. Moreover, we constructed a risk model termed HIAscore and revealed the correlation between HIAscore and biological characteristics as well as immune suppressive microenvironment of CRC patients. Among these HIA-related genes, we showed that MPI could interact with LDHA. Knocking down MPI further impaired proliferation and angiogenesis of CRC by JAK2/STAT3 signaling pathway. Thus, MPI plays an essential role in bridging hypoxia-induced angiogenesis and serves as promising therapeutic vulnerability in CRC.

Methods

Data acquisition and preprocessing

The bulk RNA-seq and clinical data of CRC patients were downloaded from TCGA project COAD and READ deposited in UCSC Xena (https://xenabrowser.net/datapages/). The expression matrix was transformed into transcript per million (TPM) and log normalized. Additionally, we obtained transcriptome profiles by microarray and clinical information if available of other 11 CRC cohorts from GEO, including 9 datasets GSE37892, GSE35896, GSE161158, GSE14333, GSE39582, GSE143985, GSE119409, GSE13294, and GSE170999 as exploration, and 2 datasets GSE17356 and GSE17357 as validation. The batch effects of different microarrays were eliminated using ComBat function from R package sva to obtain the GSE_merged cohort [21]. The raw count matrix of hypoxic models induced by various stimulus in different CRC cell lines and an endothelial cell line HUVEC were downloaded from GSE158632, GSE200204, and GSE89831. Data of chromatin immunoprecipitation-sequencing (ChIP-seq) against HIF1A and HIF2A in HCT116 were acquired from GSE200203. The single cell RNA-sequencing (scRNA-seq) data from 10 untreated CRC patients was accessed from GSE205506. Spatial transcriptome was obtained from an open-resource platform (10.5281/zenodo.7551712).

Functional enrichment analysis

We utilized both single sample gene set enrichment analysis (ssGSEA) from R package GSVA and GSEA from R package clusterProfiler to compare the signature activities between hypoxic and normoxic conditions of cells [22, 23]. Differentially expressed gene (DEG) analysis was conducted by R package DESeq2 and genes were ordered by log2 fold changes [24]. Genes in hallmarks of cancer and signature pathways were retrieved from the Molecular Signatures Database (MsigDB) and Kyoto Encyclopedia of Genes and Genomes (KEGG), respectively. Functional enrichment analyses focusing on Gene Ontology (GO) and KEGG pathways were performed using R package clusterProfiler and Metascape [23, 25]. For gene enrichment analyses, FDR correction using adjusted p values were applied.

Visualization of ChIP-seq results

Generation of peak files and bw tracks of HIF1A/HIF2A were described in the original paper [26]. We used “bedtools intersect” and “bigWigMerge” to create consensus overlapping peak regions and merged bw tracks from biological replicates, respectively [27]. Heatmap of HIF1A/HIF2A binding density across peaks was generated by “computeMatrix reference-point” and “plotHeatmap” from deepTools [28]. We annotated the peak regions to their nearest genes referred to human genome hg19 with R package ChIPseeker [29]. Promoters were defined as 3000 bp up- and down-stream distance from transcription start site (TSS). Genome track of binding enrichment was visualized by IGV.

Identifying HIA-related genes

We integrated 3 methods to identify potential HIA-related genes: (1) predicted to be transcriptional regulated by hypoxia induced factors (either HIF1A or HIF2A). Therefore, we include genes whose sequence contains hypoxia response elements and shows enrichment of either HIF1A or HIF2A by ChIP-seq. (2) increased mRNA expression in hypoxic condition compared to normoxia. We achieved this by performing DEG analysis using R package DESeq2 and the raw count matrix from GSE200204 were used as input. Genes with log2 fold change > 1 and adjusted p value < 0.05 were regarded significance. (3) potentially involved in angiogenesis. Here, we utilized our previously proposed SRGA method [30]. Briefly, we started by calculating the partial correlation between expression of one gene with the other genes considering tumor purity. Then we employed GSEA method to measure the score value of the gene corresponding to each one of the seven angiogenesis signatures obtained from MsigDB. Score value was calculated by -log10(p) * NES. We considered genes with absolute score value > 0.95 and false discover rate (FDR) < 0.05 as angiogenesis-related ones to optimize the identification of biologically relevant enrichment in this specific context. As the result, twelve genes overlapped in 3 methods were identified as HIA-related genes.

Genome-wide screening for potential targets of transcription factors (TFs)

Among the 12 HIA-related genes, NR4A2, FOXO3, and MAFK were TFs. To further explore the regulatory networks governed by HIA, we accessed JASPAR (https://jaspar.genereg.net/) to obtain the binding motif of NR4A2 (MA0160.2), FOXO3 (MA0157.3), and MAFK (MA0496.1) [31]. Subsequently, we leveraged FIMO from MEME Suite to screen potential binding sites of these TFs [32]. These binding sites were annotated to their nearest genes as described above. These genes were identified as TF binding targets. We constructed the TF-target genes network to show their prospective role in connecting hypoxia and angiogenesis.

Consensus clustering and identification of prognostic CRC subtypes

Based on the expression of 12 HIA-related genes, we applied “ConsensusClusterPlus” for tuning cluster numbers from two to six with parameters “reps = 1000, pItem = 0.8, pFeature = 1”, distance between two sample was calculated by Euclidean [33]. We chose 3 for the optimal clustering number because the delta area and consensus cumulative distribution function (CDF) increased slowly after k = 3, and patients assigned to each subtype showed distinct clinical and biological features. The prognostic value of different CRC subtypes was examined using Kaplan-Meier method with log-rank test implemented in R package survival and survminer. Multivariate Cox regression was applied by R package forestplot.

Construction and evaluation of HIAscore

To further exploit the prognostic value of HIA-related genes, we quantify the HIA pattern of individual patient from GSE_merged by constructing a scoring system termed as HIAscore. First, we identified the subtype-related DEGs using R package limma and 227 overlapped genes among three subtypes were extracted [34]. Next, we performed univariate Cox regression analysis for each overlapped gene using R package survival to determine prognosis-related genes. These genes with p value < 0.05 were regarded as significant. Then, we conducted principal component analysis (PCA) based on the expression of 56 prognosis-related genes and both principal component 1 and 2 were used for calculating the HIAscore, which was defined as: , where is the expression of prognosis-related genes. Furthermore, we investigate the prognostic value of HIAscore in both GSE_merged and TCGA cohorts. This involved in clinical information of patients, including overall survival, disease-free survival, and the distribution of HIAscore in the tumor stages, mismatch repair (MMR) status, whether relapse, and consensus molecular subtypes (CMS). The CMS of CRC patients were predicted by calculating correlations between samples and centroids from R package CMSclassifier [3].

Reanalyzing scRNA-seq data

The raw UMI count matrix was imported into R package Seurat [35]. Low quality data were filtered out if genes expressed in fewer than three cells, cells had fewer than 500 genes or more than 5000 genes expressed, or cells detected fewer than 400 unique molecular identifiers (UMI) or more than 25,000 UMI. Filtered data were normalized by function “NormalizeData” with default parameters. The top 2000 highly variable genes were generated using function “FindVariableFeatures” with “vst” method. Next, we performed PCA for dimension reduction, and the resulting first 30 PCs were used as input for Uniform Manifold Approximation and Projection (UMAP) algorithm. Cell clusters were identified with “FindNeighbors” and “FindClusters” functions at a resolution of 1. Cell types were annotated based on canonical markers. For sub-clustering of epithelial cells, we extracted the subset of epithelial cell and repeated analyses mentioned above. Each cell was assigned a HIAscore as described above. Cells in Subcluster 1 and 11 had lowest HIAscore while Subcluster 5 and 15 the highest. They were defined as HIAscore-Low and HIAscore-High, respectively. DEGs in the two subclusters were identified using function “FindMarkers” with Wilcoxon rank-sum tests. Functional enrichment analyses were performed using R package clusterProfiler [23]. The dynamic transition from HIAscore-low to -high epithelial cells was inferred by pseudotime analysis implemented in R package monocle2 [36]. The intercellular communication networks with ligand-receptor signal patterns were inferred with R package CellChat with default parameters [37].

Estimating of tumor immune infiltrating

We utilized R package ESTIMATE to measure the enrichment of immune score, stromal score, Tumor purity, and ESTIMATE score for each patient [38]. The infiltrating of each immune cell type in the tumor microenvironment (TME) was quantified using 28 signatures retrieved from TISIDB by ssGSEA [39]. We also estimate the enrichment of immune exhaustion, suppression, and exclusion in the TME with R package IOBR [40]. To identify the spatial location of HIAscore and cells, we used R package Seurat to analyze the gene-spot matrix of CRC samples. Spots with number of UMI counts below 500 or above 45,000, and containing more than 50% mitochondrial gene expression, were filtered out. Data normalization, dimensional reduction, and clustering of spots were performed using SCTransform, RunPCA, RunUMAP, FindNeighbors and FindClusters with default parameters, respectively. We applied R package SPOTlight to deconvolute spots from spatial transcriptome using annotated cell types from scRNA-seq and predict their location at CRC tissues [41].

Protein-protein interaction analyses

Potential protein-protein interactions (PPI) among HIA-related genes were searched from STRING (https://cn.string-db.org/) and GeneMANIA (http://genemania.org/) databases. The protein structure of MPI or MPI (D43A/N214D/Q243K) and LDHA was predicted by AlphaFold (https://alphafold.ebi.ac.uk/) using amino acid sequences retrieved from Uniport (https://www.uniprot.org/). The docking complexes were input into PyMOL for visualizing the docking complexes and identifying polar contact.

Cell culture

The colon cancer cell line HCT116 and human umbilical vein endothelial cells (HUVECs) were purchased from the cell bank of the Chinese Academy of Science (Shanghai, China) and Procell (Wuhan, China), respectively. HCT116 was cultured in DMEM (Cytiva, USA) plus 10% fetal bovine serum (Excell Bio, HK) and 1% penicillin/streptomycin (Beyotime, China). To induce hypoxia in the hypoxia group, 125 or 250 µM cobalt chloride (CoCl₂) was used to treat cells for 12–24 h. HUVECs were cultured in Endothelial cell medium containing 5% FBS and 1% endothelial cell growth factor (all purchased from Procell). All cells were cultivated in a humidified incubator at 37 °C with 5% CO₂.

Plasmid vector construction and cell transfection

MPI knockdown was performed using the short hairpin RNA (shRNA) system. The shRNA oligos for vector construction were purchased from Tsingke Biological Technology Co., Ltd. The detail information of these oligos were listed in Supplementary Table S1. Next, the recombinant pLKO.1-shMPI or shHIF-1α vector were co-transfected with packing plasmids psPAX2 and PMD2.G into 293T cells. The viral supernatant was harvested to infect HCT116 cells. Infected HCT116 cells were selected with 2.5 µg/mL puromycin for 2 days before performing the subsequent experiments.

Colony formation assay

HCT116 cells were digested with trypsin to obtain a single cell suspension. 3 × 10³ cells were seeded onto a 6-well plate with different for 2 weeks. The colonies were carefully washed twice with phosphate-buffered saline (PBS). Then, the colonies were fixed with 4% paraformaldehyde for 30 min and stained using crystal violet for 30 min.

Tube formation assays

50 µL of Matrigel matrix (Corning, NY, USA) was plated in 96-well plate and incubated at 37 °C for 30 min to allow the Matrigel to polymerize. HUVECs were pre-treated with HCT116 supernatant for two days and 2.5 × 10⁴ treated HUVECs were seeded on the Matrigel-coated wells. Then the plate was incubated at 37 °C for 12 h. Tube formation was observed with microscope after stained with Calcein AM (Beyotime, Shanghai, China). Total number of junctions were calculated to assess the tube formation ability.

RT-qPCR

Total RNA from cell was isolated using a Cell Total RNA Isolation Kit (Foregene, Chengdu, China). Reverse transcription was conducted using the HiScript III RT SuperMix (Vazyme, China) according to the manufacturer’s protocol. RT-qPCR was performed using a CFX96 Real-Time PCR System (Bio-Rad, USA) with ChamQ SYBR qPCR Master Mix (Vazyme, China). The relative expression levels of target genes were normalized to β-actin. The primers for RT-qPCR used in this study were list in Supplementary Table S2.

Western blot

The cell precipitates were suspended in RIPA lysis buffer and incubated on ice for 30 min. Then the total protein concentration was measured by a BCA Protein Assay Kit. Appropriate amount of protein loading buffer was added to the protein sample and were then heated to 99 °C for 10 min. 20 µg total protein loaded in each lane was separated on SDS-polyacrylamide gels and then transferred onto polyvinylidene difluoride (PVDF) membranes (Bio-Rad, USA). Then, after blocking with 5% skim milk powder for 1 h at room temperature, the membranes were cut and incubated in primary antibody overnight at 4 °C. Proteins were detected with the antibodies as follows: LDHA (19987-1-AP, proteinTech, China, 1:1000 dilution), MPI (14234-1-AP, proteinTech, China, 1:1000, dilution), JAK2 (ab108596, Abcam, UK, 1:1000 dilution), STAT3 (ab68153, Abcam, UK, 1:1000 dilution), p-JAK2 (ab32101, Abcam, UK, 1:1000 dilution), p-STAT3 (ab32143, Abcam, UK, 1:1000 dilution), GAPDH (TA-08, ZSGB-BIO, China, 1:1000 dilution), β-actin (TA-09, ZSGB-BIO, China, 1:1000 dilution). GAPDH and β-actin were used as the sample loading control. After washed for 5 times in 1 × TBST, membranes were incubated with a secondary antibody for 2 h at room temperature. Another 5 times washing in 1 × PBS, the immunostaining intensity of each protein band was quantified using a Bio-Rad ChemiDoc XRS system (Bio-Rad, USA).

Co-immunoprecipitation (Co-IP)

Co-IP was performed as previously described [42]. Briefly, HCT116 cells were washed twice with PBS and lysed with lysis buffer for 30 min. After removing cell debris by centrifugation, the supernatant was incubated with specific antibodies in a rotator overnight. Then, Protein A/G Magnetic Beads (MedChemExpress, Shanghai, China) were added to bind above complex for 2 h at in a rotator. After 5 times’ wash in lysis buffer, protein A/G beads were diluted with 1× protein loading buffer and heated to 99 °C for 10 min. Last, samples were detected by western blot.

Mouse model

All animal experimental procedures were in accordance with protocols approved by the Experimental Animal Ethics Committee of West China hospital, Sichuan University (Ethics record number 20220107010). Male BALB/c mice (4–6 weeks) were purchased from HUAFUKANG (Beijing, China) and housed in a pathogen free animal facility. We subcutaneously injected HCT116 cells (5 × 10⁶) transfected with sh-scramble or sh-MPI (the mixture of three knockdown groups) into the flanks of mice. Tumor size was measured every three days after one week using a caliper. Tumor volume was calculated as length × width²/2. Then, mice were sacrificed 16 days after modeling by cervical dislocation and subcutaneous tumor xenografts were photographed. In addition, Immunohistochemistry (IHC) staining for Ki67 and CD31 was performed.

Statistical analysis

R software 4.0.2 (https://www.R-project.org/) and GraphPad 8.0 were used for statistical analyses. Wilcoxon test was used for comparison between two groups. Analysis of variance (ANOVA) was used for calculating difference among three groups. Log-rank test in R package survminer was used for survival analysis. The relationship between hypoxia and other cancer hallmarks were calculated by Pearson correlation analysis. All experiments were performed by triplicate and presented as the mean ± SD. Two-sided Student’s t-test were used for statical analyses. Results with p value < 0.05 was regarded as statistically significance.

Results

Correlation between hypoxia and angiogenesis in CRC

Solid tumor cells require oxygens and nutrients to sustain proliferation. In hypoxic condition, CRC secrete a plethora of factors to stimulate distal endothelial cells to form new blood vessel and promote tumor progression. This study aims to unravel the transcriptional regulatory network behind HIA using multi-omics analyses and experimental validations. A schematic diagram was listed in Fig. 1A. We first downloaded the expression matrix of 2292 tumor tissues from 10 clinical CRC datasets. Then, ssGSEA and Pearson correlation analysis were performed to calculate the relations among cancer hallmarks in each dataset. Expectedly, we found that hypoxia was positively correlated with several hallmarks, including P53, IL6/JAK/STAT3, TGF-beta signaling pathways, and angiogenesis (Fig. 1B and Fig. S1A). Angiogenesis is the process of endothelial cells sprouting to induce new blood vessels formation and responsive to hypoxia. Indeed, angiogenesis was enriched in hypoxic HUVEC cells (Fig. 1C and Fig. S1B). We noticed that the enrichment of angiogenesis was also escalated in different CRC cell lines after various hypoxic stimulation (Fig. S1C-D). To further investigate the potential correlation between hypoxia and cancer hallmarks in other cell types, we downloaded and processed a CRC scRNA-seq dataset, then annotated 6 major cell types based on their marker genes (Fig. 1D and Fig. S1E). Pearson correlation analysis results confirmed that hypoxia was positively correlated with angiogenesis in endothelial cells, fibroblasts, epithelial cells, and myeloid cells (Fig. 1E and Fig. S1F). These results suggest that hypoxic treatment would induce the expression of angiogenesis-related genes, which was common in many cell types, while only endothelial cells display angiogenesis phenotype.

Fig. 1.

Hypoxia is highly correlated with angiogenesis in multiple datasets. (A) Schematic diagram of this study. (B) Heatmap showing the Pearson correlation of ssGSEA score between hypoxia and other cancer hallmarks in multiple CRC datasets. (C) GSEA result showing the enrichment of angiogenesis by comparing DEGs between hypoxic and normoxic conditions. (D) UMAP plot showing the distribution of 6 main cell types in CRC scRNA-seq data. T/I/NK cells represent T cells/innate lymphocytes/NK cells. (E) Heatmap showing the Pearson correlation of ssGSEA score between hypoxia and other cancer hallmarks of the 6 main cell types in CRC scRNA-seq data

Identification of HIA-related genes

We sought to investigate regulatory genes underpinning HIA by comprehensive bioinformatic analyses. Since HIF1A and HIF2A are highly conserved transcription factors directly mediating the downstream effects of hypoxia, we first obtained their binding regions on chromatin using publicly available ChIP-seq data. In total, there were 2528 binding peaks of HIF1A and most of them were enriched in promoter regions of genes, while 350 binding peaks were observed for HIF2A, which mainly located at protomer and distal intergenic regions (Fig. 2A and Fig. S2A, Supplementary Data 1–2). We annotated these peaks to their nearest genes and kept only autosomal genes showing HIFs binding at their promoters. As a result, 2015 genes remained. Since hypoxia generally promotes gene transcription, especially for those involved in angiogenesis, we next compared the differentially expressed genes between hypoxic and normoxic cultured HCT116, and identified 941 up-regulated genes (Fig. 2B, Supplementary Data 3). We then applied signature-related gene analysis (SRGA) method, a computational framework we previously developed, to explore potential angiogenesis-related genes. Using transcriptome profiles from TCGA-CRC and angiogenesis signatures, 911 unique genes were proposed to be angiogenesis-related genes (Fig. 2C, Supplementary Data 4). We defined the 12 genes concordantly appeared in the three gene list as the HIA-related genes hereafter (Fig. 2D). IGV track visualizations confirmed the specific bindings of HIFs at their promoter regions (Fig. S2B-M). Gene enrichment analysis annotated that HIFs were involved in cellular response to oxidative stress as well as VEGFA-VEGFR2 signaling pathway (Fig. 2E). We also retrieved the downstream targets of 3 transcriptional factors from HIA-related genes, and constructed a regulatory network showing the potential role of HIAs in bridging hypoxia and angiogenesis (Fig. 2F).

Fig. 2.

Identification of HIA-related genes and their regulatory networks in CRC. (A) Heatmap (left) and Pie graph (right) showing the genome-wide ChIP-seq enrichment, and distribution and percentage of HIF1A and HIF2A on their binding peaks, respectively. (B) Volcano plot showing the DEGs between hypoxic and normoxic conditions. Red dots represent significant up-regulated genes with log2FC > 1 and adjusted p value < 0.05. (C) Bar chart showing the number of genes relevant to 8 angiogenesis signatures. (D) Venn plot showing the overlapping genes among 3 methods. (E) Metascape network showing similarities of enriched terms of the 12 HIA-related genes colored by clusters. (F) Network plot showing the role of 12 HIA-related genes in connecting hypoxia and angiogenesis

Validation of HIA-related genes

We next examined the expression of HIA-related genes in experimental conditions. HCT116 cells were first treated with CoCl₂ at various concentrations and durations to confirm its role in inducing hypoxia (Fig. S3). Then, RT-qPCR and western blot were performed to confirm the levels of HIF-1α expression. The results revealed that both the mRNA and protein levels of HIF-1α expression were significantly increased under hypoxic condition compared to normoxia, while reduced after knocking down HIF-1α (Fig. 3A-B). Subsequently, RT-qPCR was conducted to examine the expression of HIA-related genes. As expected, these 12 genes were significantly up-regulated under hypoxic condition. Upon HIF-1α knockdown, the expression of these genes was markedly reduced, with the exception of TBC1D22A (Fig. 3C-N). Then, the supernatant of HCT116 was harvested and added to HUVECs. Supernatant from HCT116 treated with CoCl₂ induced significantly more tube formation of HUVECs than that from the normoxic one. This effect was repressed by knocking down HIF-1α (Fig. 3O-P). These results suggest that HIA-related genes are indeed up-regulated by HIFs under hypoxic condition.

Fig. 3.

Validation of HIA-related genes. (A) The expression of HIF-1α mRNA in HCT116 with or without CoCl₂ treatment and HIF-1α knockdown was validated by RT-qPCR. β-actin was used as internal control. (B) The expression of HIF-1α protein in HCT116 with or without CoCl₂ treatment and HIF-1α knockdown was assessed by western blot. (C-N) RT-qPCR was applied to validate the mRNA expression of 12 HIA-related genes including ANKZF1, EPB41L1, FOXO3, HSPG2, LDHA, MAFK, MPI, NR4A2, PDLIM1, RCOR2, RRAGD, TBC1D22A. β-actin was used as internal control. (O) Representative images of tube formation in HUVECs were captured after stained with Calcein-AM. (P) Quantification of junction numbers in HUVECs

Identification of three subtypes with distinct biological characteristics based on HIA-related genes

To further explore the clinical application value of HIA-related genes, we applied consensus clustering method based on their expression and identified 3 CRC subtypes (Fig. S4A). Among the 1686 CRC patients, 672 were classified as Subtype1, while 558 and 456 patients were stratified into Subtype2 and Subtype3, respectively. Principle component analysis showed clear separation among these subtypes (Fig. 4A). There were no obvious differences in gender and T stage distribution across the 3 subtypes (Fig. 4B). However, Subtype2 and Subtype3 comprised higher proportion of senior patients (age > 65). Notably, patients in Subtype2 exhibited greater incidence of lymphatic metastasis (N1-N3), distal metastasis (M1), and advanced clinical stage (Stage III + Stage IV) compared to the other 2 molecular subtypes (Fig. 4B). Thus, patients assigned to Subtype2 had the worst overall and disease-free survival (Fig. 4B-C and Fig. S4B). We noticed that the expression of HIA-related genes was varying among the 3 subtypes, implying different hypoxia and angiogenesis patterns in patients (Fig. S4C). Specifically, Subtype2 exhibited more aggressive tumor hallmark characteristics comprising transcriptional misregulation and increased activities in several signaling pathways, such as PI3K-AKT, JAK-STAT, and TGF-Beta (Fig. 4D). Besides, GO terms summarized as extracellular matrix remodeling and structural organization, as well as immune cell trafficking and motility were features of Subtypes2, demonstrating aberrant TME involved in tissue fibrosis and immune cell recruitment (Fig. 4E). Metabolic pathways were hyperactively enriched in Subtype3, leading to disturbance of cell cycle regulation and mitosis, such as small molecule catabolic process, mitotic nuclear division, and cell cycle checkpoint signaling (Fig. 4D-E). Subtype1 were enriched in mixed features, including digestive system, DNA repair, signaling transduction, and metabolism (Fig. 4D-E). As for tumor driver gene mutation, we found that patients in Subtype1 had higher frequency of TP53 mutation and lower frequency of BRAF mutation compared to the other 2 subtypes, while Subtype3 had the opposite trend (Fig. 4F). Both Subtype2 and Subtype3 had similar KRAS mutation frequency (about 40%), slightly higher than that in Subtype1 (Fig. 4F). In summary, HIA-related genes successfully classified CRC patients into 3 molecular subtypes with different prognosis and biological characteristics.

Fig. 4.

Identification and characteristics of 3 CRC subtypes based on HIA genes. (A) PCA projection of samples from GSE_merged cohort colored by consensus clustering results. (B) Pie charts for the distribution of different clinical characteristics in the 3 subtypes. (C) Kaplan-Meier plot showing the overall survival (top) and disease-free survival (bottom) of 3 CRC subtypes in GSE39582. (D) Heatmap showing different ssGSEA enrichment of KEGG pathways among three subtypes in GSE_merged cohort. (E) Top10 GO enrichment analyses of marker genes from 3 subtypes. (F) Mutation rate of BRAF, KRAS, and TP53 among 3 subtypes. The p value is calculated by Chi-squared test

Development and evaluation of the prognostic model

To better depict the role of HIAs in shaping prognosis of patients, we determined 227 overlapping differentially expressed genes among the 3 subtypes by limma package (Fig. 5A). Then, we performed univariate Cox regression analysis to select prognostic genes for overall survival. As a result, forty-four and 12 genes were significantly associated with favorable and adverse prognosis (Fig. 5B and Supplementary Data 5). We subsequently constructed a scoring system termed HIAscore to quantify the HIA pattern at the individual patient level. Based on the optimal cutoff using surv_cutpoint function in the survminer R package, we classified patients into high and low groups. High HIAscore group exhibited significant enrichment in both hypoxia and angiogenesis signatures retrieved from MSigDB. Consistent results were observed in GSE and TCGA-CRC cohorts (Fig. 5C and Fig. S5A). In addition, survival analyses revealed that patients with higher HIAscore had significantly shorter overall and disease-free survival time compared to those with lower HIAscore (Fig. 5D and Fig. S5B). To assess whether HIAscore could serve as an independent prognostic factor, we also performed multivariate Cox regression analysis using clinical variables including age, gender, and TNM stage. HIAscore emerged as a robust and independent risk factor for patient outcomes in GSE39582 (Fig. 5E, HR = 1.5, p = 0.013) and TCGA-CRC (Fig. S5C, HR = 1.6, p = 0.029). We noticed that HIAscore was increased from early stage to advanced CRC stage (Fig. 5F and Fig. S5D), and high HIAscore group contains more patients diagnosed with advanced stage disease (Fig. 5G and Fig. S5E). Besides, patients with proficient mismatch repair had higher HIAscore compared to the deficient ones (Fig. 5H and Fig. S5F). We also observed significantly higher HIAscore in relapsed tumors (Fig. 5I and Fig. S5G). We further examined the association between HIAscore and CMS subtypes. HIAscore varied considerably among subtypes, with CMS4 subtype exhibiting the highest score and CMS3 the lowest (Fig. 5J). Likewise, the distribution of CMS subtypes was significantly different between the high and low HIAscore groups. The CMS4 subtype was more frequent in the high HIAscore group, while the CMS3 subtype was predominant in the low group (Fig. 5K). These association were consistently validated in the TCGA-CRC cohort (Fig. S5H-I). These results suggest that HIAscore reflects the biological characteristics of CRC patients, and may serve as a useful biomarker for prognosis and molecular classification.

Fig. 5.

Construction and association between HIAscore signature and clinical features. (A) Upset plot showing the overlapped differentially expressed genes among 3 subtypes. (B) Dot plot showing the cox hazard ratio of each overlapped gene using GSE39582 cohort. Red and blue dots represent protective and risk genes with hazard ratio < 1 or > 1, and p value < 0.05, respectively. (C) Heatmap showing the relationship between normalized HIAscore and ssGSEA enrichment of angiogenesis and hypoxia-related pathways using GSE_merged cohort. Samples were ranked by ascending HIAscore. (D) Kaplan-Meier plot showing the overall survival of two HIAscore groups in GSE39582, TCGA, GSE17536, and GSE17537 cohorts. (E) Forest plot showing the multivariate Cox regression results including age, gender, stage, and HIAscore in GSE39582 cohort. (F) Boxplot showing the normalized HIAscore among different stages in GSE39582 cohort. The p value is calculated by Wilcoxon tests. (G) The proportion of patients with different stages from two HIAscore groups in the GSE39582 cohort. (H-I) Boxplot showing the normalized HIAscore between different MMR status (H) and relapse tumors (I) in GSE39582 cohort. The p value is calculated by Wilcoxon tests. (J) Boxplot showing the normalized HIAscore among different CMS in GSE_merged cohort. The p value is calculated by Wilcoxon tests. (K) The proportion of patients with different CMS from two HIAscore groups in the GSE_merged cohort

HIAscore correlates with immune suppressive TME

To investigate the relationship between HIAscore and TME, we utilized ESTIMATE algorithm and identified significant differences between the two HIAscore groups. Specifically, immune score and tumor purity were higher in low HIAscore group, while stromal score and ESTIMATE score were increased in high group (Fig. 6A and Fig. S6A). Immune infiltrating cells are major components of TME and play dual roles during tumor progression. We calculated the relative proportion of 28 immune cells in each CRC sample based on curated signatures from tumor-immune system interactions database (TISIDB). The results revealed that the enrichment score of activated immune cells, including B cells, CD4/8 T cells were higher in low HIAscore group (Fig. 6B and Fig. S6B). In particular, the high HIAscore group was characterized by relatively higher proportion of immunosuppressive cells, namely regulatory T cell (Treg), nature killer T cell, and Plasmacytoid dendritic cell (pDC). However, the enrichment scores of immune cells exerting anti-tumor activities, including nature killer cell and memory B cell, were comparatively higher in high HIAscore group at the same time (Fig. 6B and Fig. S6B). To further depict the immune feature of patients, we measured enrichment score of published signatures involved in immune exhaustion, immune suppression, and immune exclusion from immuno-oncology biological research (IOBR). We noticed that cancer-associated fibroblasts (CAFs), as well as activation of EMT, WNT, and TGFb signaling pathways were consistently enriched in high HIAscore group (Fig. 6C-E and Fig. S6C-E). Using SPOTlight deconvolution algorithm, we projected the location of epithelial cells, fibroblasts, and T/I/NK cells (T cells/innate lymphocytes/NK cells) from scRNA-seq to the spatial transcriptome. Interestingly, HIAscore signature was primarily within epithelial cells region, while fibroblasts and T/I/NK cells were largely excluded from these areas (Fig. 6F and Fig. S6F). Collectively, these findings suggested that high HIAscore may indicate “cold tumor” phenotype characterized by immune sequestration.

Fig. 6.

Comparison of tumor immune microenvironment between HIAscore high and low groups. (A) Violin plot showing the calculated value of ImmuneScore, StromalScore, TumorPurity, and ESTIMATEScore between two HIAscore groups in the GSE_merged cohort. The p value is calculated by Wilcoxon test. (B) Different ssGSEA enrichment of immune cell infiltrating between two HIAscore groups in the GSE_merged cohort. (C-E) Boxplot showing ssGSEA enrichment of immune exclusion (C), immune exhaustion (D), and immune suppression (E) signatures between two HIAscore groups in GSE_merged cohort. The p value is calculated by Wilcoxon test. The top, middle, and bottom lines of the boxes represent the 75th, median, and 25th percentile values, respectively. *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001, ns: not significant. (F) Spatial feature plot displaying predicted distribution and expression of HIAscore, epithelial cells, fibroblasts, and T/I/NK cells

Analyzing HIAscore at single cell level

We further investigated the expression of HIA-related genes and HIAscore at single cell resolution. We found that all 12 HIA-related genes were expressed in epithelial cells. LDHA was highly expressed in all 6 cell types, while the expression of RCOR2 was relatively low (Fig. S7A). Subsequently, we calculated the HIAscore for each cell and observed heterogeneous in epithelial cells (Fig. S7B). This enthuse us most because HIAscore signature was mainly colocalized within epithelial cells region spatially (Fig. 6F and Fig. S6F). We then subclustered epithelial cells into 26 groups and ranked them by median HIAscore in ascending order (Fig. S7C-D). Subcluster 1 and 11 had relatively low HIAscore, whereas Subcluster 5 and 15 showed elevated scores (Fig. S7D). We defined them as HIAscoreLow and High group, respectively. Functional enrichment analysis based on DEGs between the two groups showed that cells with high HIAscore were enriched in several oxidative-related pathways, such as oxidative phosphorylation and cellular response to hypoxia (Fig. 7A). On the other hand, cells in the HIAscoreLow group appeared to be metabolic active and enriched in pathways including mineral absorption and cellular lipid catabolic process (Fig. 7A). Pseudotime analysis identified two trajectory branches, reflecting the dynamic cellular transition from HIAscoreLow to HIAscoreHigh states (Fig. 7B). Consistently, the enrichment scores of both hypoxia and angiogenesis gradually increased along the pseudotime trajectory, accompanied by up-regulation of the 12 HIA-related genes (Fig. 7C-D). We performed cell-cell interaction analysis to infer the intercellular communication among HIAscoreHigh and Low epithelial cells, fibroblasts, and T/I/NK cells (Supplementary Data 6). We observed higher activity of MDK-mediated MK signaling pathway in HIAscoreLow compared to HIAscoreHigh epithelial cells, a pathway known to drive immune cell chemotaxis [43] (Fig. 7E-F). Moreover, HIAscoreHigh epithelial cells served as the core sender in several cell communication networks, particularly LAMININ and MHC-I signaling pathways (Fig. 7E-F). LAMININ signaling pathway had been shown to promote angiogenesis [44]. The ligand-receptor pair HLA-C-KIR2DL3 within MHC-I signaling pathway network is known to inhibit NK cells activity, thereby preventing HIAscoreHigh epithelial cells lysis [45]. These results provided preliminary molecular evidence into the mechanisms by which HIAscoreHigh epithelial cells escape immune surveillance.

Fig. 7.

HIAscore distribution and biological significance in CRC at single cell level. (A) KEGG pathway enrichment of marker genes from HIAscore high (left) and low (right) epithelial cell subclusters. (B) Construction and visualization of pseudotime scores from HIAscoreLow to HIAscoreHigh. (C-D) ssGSEA scores and expression profiles of hallmark signatures (C) and HIA-related genes (D) along different pseudotime trajectories. (E) Ligand–receptor communication networks of MHC-I, LAMININ, and MK signaling pathways between HIAscore-High and -Low epithelial cells and fibroblasts, T/I/NK cells predicted by CellChat. (F) Circle plot showing the inferred interaction strength of MHC-I, LAMININ, and MK signaling networks among different cell types

MPI interacts with LDHA and affects proliferation and angiogenesis of CRC through JAK2/STAT3 signaling pathway

To gain mechanistic insights into the functions of HIA-related genes, we performed correlation and protein-protein interaction (PPI) analyses. The expression of MPI was positively correlated with LDHA in CRC (Fig. S8A). Moreover, LDHA was predicted to interact with MPI in both STRING and GeneMANIA databases (Fig. S8B). Given the well-established role of metabolic reprogramming in angiogenesis [46, 47], we focused on these two metabolism-related genes. In silico molecular docking predicted a direct binding interaction between MPI and LDHA (Fig. S8C). While this interaction was abolished in MPI (D43A/N214D/Q243K) point mutation (Fig. S8D). Co-IP assays experimentally confirmed the interaction between LDHA and MPI (Fig. S8E). Specifically, LDHA has been shown to promote angiogenesis by regulating surface vimentin expression and enhancing VEGFR phosphorylation in endothelial cells [48–50]. However, the function of MPI in regulating angiogenesis remains incompletely understood. We first knocked down MPI in HCT116 cell line (Fig. 8A). Colony formation assays revealed that MPI knockdown led to impaired proliferation (Fig. 8B-C). Tube formation assays showed that MPI knockdown can suppress tube-forming ability of HUVECs (Fig. 8D-E). Furthermore, we aimed to explore the downstream pathways linked with MPI and angiogenesis. CRC patients in 10 independent datasets were stratified into MPI high and low groups based on its median expression. GSEA revealed significant enrichment of JAK2/STAT3 pathway in most MPI high groups (Fig. 8F and Fig. S9). Western blots were conducted to confirm this finding (Fig. 8G). Next, to further explore the function of MPI in vivo, subcutaneous xenograft tumor models were established. MPI knockdown significantly inhibited tumor proliferation in vivo (Fig. 8H-I). The results of Ki67 (proliferation marker) IHC staining further supported this finding (Fig. 8J-K). Meanwhile, the results of CD31 (endothelial cell marker) IHC staining suggested that MPI knockdown could decrease neovascular density (Fig. 8J and L). Collectively, these findings suggested that MPI interacts with LDHA and promoting proliferation and angiogenesis in CRC through JAK2/STAT3 signaling pathway.

Fig. 8.

Knocking down MPI inhibits proliferation and angiogenesis of CRC through JAK2/STAT3 signaling pathway. (A) Validation of MPI knockdown by western blot. (B) Representative images of colony formation assays in HCT116 after MPI knockdown. (C) Quantification of colony numbers in panel B. (D) Representative images of tube formation in HUVECs were captured after stained with Calcein-AM. (E) Quantification of junction numbers in panel D. (F) Normalized enrichment score of IL6-JAK-STAT3 signaling pathway calculated by GSEA analyses by comparing MPI high and low groups in multiple CRC datasets. (G) The expression of JAK2/STAT3 signaling pathway-related proteins was detected by western blot. (H) Subcutaneous xenograft tumors harvested on day 16 from BALB/c nude mice. (I) Volume of subcutaneous xenograft tumors measured every three days. (J) Representative images of IHC staining for Ki67 and CD31 in subcutaneous xenograft tumors. (K) Relative intensity of Ki67 was calculated. (L) Mean CD31 positive microvascular density was calculated. Each experiment was conducted three times. Data was presented as mean ± standard deviation (SD). *: p < 0.05, **: p < 0.01

Discussion

Both hypoxia and angiogenesis are leading features of malignancies and influence the proliferation and progression of tumor cells by remodeling TME [11, 13, 18, 19]. While most studies focused on these hallmarks separately, their interplay and combined effects in CRC are not fully understood. The identification of key transcription factors HIFs and their downstream targets, including VEGF, PDGF, and others, has advanced our understanding towards angiogenesis. These findings have promoted the progress of scientific research and clinical application. Yet, one of the remaining challenges lies in deciphering the profound and complex transcriptional regulatory network that connects hypoxia and angiogenesis. Biomarkers derived from this process that can reliably predict the prognosis of CRC patients necessitate further investigation. Moreover, delving into the molecular mechanisms behind HIA-related subtypes of CRC patients is of critical importance. Thus, exploiting and targeting key gene to develop efficient therapeutic strategies for cancer management is possible.

In this study, we first performed comprehensive bioinformatic analyses to elucidate the regulatory mechanisms underlying hypoxia and angiogenesis in CRC. We showed that hypoxia was positively correlated with angiogenesis across multiple cell types, implying a broad regulatory association. But the phenotypic manifestation of angiogenesis appeared to be mainly restricted to endothelial cells. HIFs are master transcription factors that mediate hypoxia-related biological processes. Therefore, we collected HIF1A and HIF2A ChIP-seq performed in HCT116 to identify their direct transcriptional targets, enabling genome-wide screening of potential binding sites and downstream genes. Given that activate transcription generally leads to elevated mRNA expression, we also compared RNA-seq data of HCT116 cultured under hypoxic and normoxic condition to identify genes that were transcriptionally upregulated in response to hypoxia. To further determine the angiogenesis-related genes, we applied SRGA, which integrates correlation analysis with predefined angiogenesis pathways. The SRGA method effectively identifies phenotype-related genes using large-scale transcriptomics and prior biological knowledges, proving both practicality and credibility [30]. Overall, twelve overlapped genes, including 3 transcription factors, were selected by the integrative analyses and termed HIA-related genes. These genes were enriched in VEGF and in pathways associated with cellular response to stimulus and oxidative stress, which were well known mediators of hypoxia and angiogenesis [6, 7]. The regulatory network cascade hypoxia and angiogenesis were also constructed based on these genes. Using an in vitro cell culture experiment, we confirmed that expression of the 12 genes increased after hypoxic treatment.

Inter-tumor heterogeneity across patients accounts for the diversity of cellular and molecular features of TME. Deciphering such feature contributes to a better understanding of CRC subtypes, allowing tailored therapeutic strategies and benefiting patients to individually precision medicine [3]. Based on the expression of HIA-related genes, we identified three distinct molecular subtypes of CRC patients, among which Subtype2 had both the worst overall survival and disease-free survival. Comparing ssGSEA results among 3 subtypes, we showed that several signaling pathways, including FoxO, PI3K-Akt, JAK-STAT, TGF-beta, and NF-kappa B and others, which intrinsically modulate cancer development through promoting cell proliferation and inhibiting cell death [51–53], were significantly activated in Subtype2. These aberrant pathways may drive more aggressive tumor phenotype, leading to poor prognosis of CRC patients.

Subsequently, we conducted different expression analysis among 3 subtypes and identified 227 overlapped DEGs. We then constructed a scoring system called HIAscore based on signature genes associated with overall survival to quantify the HIA activity of each patient. HIAscore was significantly correlated with enrichment score of hypoxia and angiogenesis-related pathways, suggesting that it recapitulates such condition of patients well. Higher HIAscore was both a poor prognosis indicator, as measured by Kaplan-Meier model, and a risk factor for CRC patients with older age and advanced stage, as measured by multi cox regression model. Most CRC are pMMR and characterized by lower tumor mutational burden and absence of tumor-infiltrating lymphocytes, rendering them less responsive to chemotherapy and immune checkpoint inhibitors compared to dMMR tumors [54, 55]. Researchers have previously classified CRC into 4 distinct CMS [3]. CMS4 is typified as Mesenchymal and represented by stromal infiltration, TGF-beta activation, angiogenesis, and worse relapse-free and overall survival. Consistently, we found that HIAscore was higher in CRC with advanced stage, pMMR status, relapse tumors, and CMS4, suggesting HIA plays detrimental role in CRC. Additionally, we found that patients in HIAscore high group had relatively lower immune score and higher stromal score. We utilized several computational methods, including MCPcounter [56], EPIC [57], and ssGSEA-based cell type-related signatures, to estimate the abundance of different cell types in the TME. Results revealed that the estimated proportion of immunosuppressive cells, especially CAFs, was significantly increased in HIAscore high group. CAFs have been reported to be sensitive to hypoxia, and involved in the crosstalk with other cells to support cancer cell proliferation, extracellular matrix remodeling, angiogenesis, and immunosuppression. These processes ultimately contribute to drug resistance and poor clinical outcome [58, 59]. Thus, targeting activated CAFs may serve as a potential therapeutic strategy for patients with high HIAscore. Of note, these results were validated in two independent datasets, confirming their biological significance instead of technical bias. Moreover, spatial transcriptome analysis revealed a barrier between epithelial cells with high HIAscore and T/I/NK cells, suggesting that HIA sustained poor immune infiltration. Further exploration at single cell resolution revealed that HIAscore is heterogenous expressed in epithelial cells. Trajectory analyses confirmed the upregulation of 12 HIA-related genes during dynamic transition from HIAscoreLow to HIAscoreHigh. Cell-cell interaction analysis demonstrated that high HIAscore epithelial cells lacked MDK growth factor secreting, limiting immune cells recruitment [43]. This type of cells could also promote angiogenesis by transducing LAMININ signaling pathway [44]. MHC-I is a double-edge sword that activate T cells and inhibit innate immunity. The expression of MHC-I on tumor cell surface was increased under hypoxic condition, leading to higher susceptibility to cytotoxic T cells [60]. However, the spatial isolation of HIAscore high epithelial cells and T/I/NK cells limits the effect of immune elimination. Additionally, the interactions between tumor cells and NK cells through HLA-E-CD94:NKG2A and HLA-C-KIR2DL3 mediate immune evasion [61, 62]. Therefore, HIA might conduce to tumor cells immune tolerance.

Finally, we sought to understand the regulatory mechanism of HIA-related genes in angiogenesis. Both correlation and PPI analyses indicated that MPI and LDHA, two metabolism-related genes, may be synergically involved in this process. The role of LDHA in linking hypoxia and angiogenesis has been well studies [48–50]. Multiple line of evidence have proved that the promoter region of LDHA contains HRE. LDHA promotes angiogenesis by regulating surface vimentin and enhancing VEGFR phosphorylation in endothelial cells. MPI encodes the phosphomannose isomerase that catalyzes the interconversion between fructose-6-phosphate and mannose-6-phosphate. Accumulation of the latter one impairs the further metabolism of glucose and renders tumor cells more susceptible to conventional chemotherapy [63]. The combined application of mannose- and mannose-6-phosphate-targeted delivery systems with chemotherapeutics holds potential for cancer treatment [64]. Nevertheless, the role of MPI in regulating hypoxia and angiogenesis in CRC is less known. We showed that there were also HRE at the promoter region of MPI, and the expression of MPI was increased after cells exposure to hypoxic condition. Knocking down HIF-1α diminished such effect. Further experiments demonstrated that knocking down MPI inhibited the proliferation of HCT116, the tube formation of HUVEC, and the angiogenesis ability of endothelial cells using both in vitro and in vivo models. Additionally, we suggested that MPI affects angiogenesis through activating JAK2/STAT3 signaling pathways.

The present study had several limitations. First, while leveraging publicly available datasets was convenient, integration of multiple datasets inevitably introduced technical biases due to differences in sample preparation, batch effects, and detection thresholds. The incomplete clinical metadata, e.g. ethnical diversity, treatment responses, in some cohorts compromised interpretation of associations with biological features. To mitigate these issues, our in-house data collection would adhere to standardized protocols and optimized pipelines should be taken in to consideration in the future. Second, most of our findings were based on correlative analyses primarily reflecting transcriptional regulatory trends, but might not fully resolve spatial or cellular heterogeneity of CRC. Therefore, experiments including flow cytometry-based quantification and multiplex immunofluorescence, were warranted. Third, the translational potential of HIAscore in clinical practice and its generalizability across diverse patient populations was still underdeveloped, prospective validation in independent cohorts was required. Fourth, although our data provided preliminary evidence for MPI severing as a mediator between hypoxia and angiogenesis, several additional experiments should be performed to reinforce the mechanistic findings. For example, the interaction between MPI and LDHA should be validated using experimental evidence. Where was the specific binding domains or sites between these two proteins should be explored. Truncation mutants and in situ proximity ligation assays could be further performed to answer this question. Additionally, structural studies using cryo-electron microscopy or X-ray crystallography could provide deeper insights. Fifth, although MPI was reported to suppress metabolism of glucose in glycolysis, the tricarboxylic acid cycle, the pentose phosphate pathway and glycan synthesis. Inhibiting mannose metabolism sensitizes cancer cells to chemotherapy [63, 65]. It is not known what was the broad metabolic role of MPI in CRC under hypoxic condition. Sixth, how MPI influence phosphorylation of JAK2/STAT3, and the impact on their transcriptional targets is not clear. Seventh, while this study establishes MPI as a correlative biomarker of immune exclusion, in vitro and in vivo functional studies using patient-derived organoids are planned to dissect the causal mechanisms. What was the therapeutic relevance of targeting MPI alone or in combination with anti-angiogenic therapies or immune checkpoint inhibitors remains poorly understood. And if there were any other molecular mechanisms exist remained to be solved in the future.

Conclusion

In conclusion, our study shed light on the importance of hypoxia and angiogenesis in CRC. We constructed a regulatory network containing 12 genes that are transcriptionally regulated by HIFs and potentially involved in angiogenesis. The 12 genes classified CRC patients into 3 Subtypes with distinct prognosis and pathway enrichment. We further extracted signature genes based on DEGs and cox hazard model to build the HIAscore system to quantify such biological characteristics in CRC patients. We showed that high HIAscore is an unfavorable indicator for poor prognosis and aggressive phenotype of CRC that relevant to immune suppression. Finally, we demonstrated that MPI is responsive to HIFs under hypoxic condition and regulates angiogenesis by JAK2/STAT3 signaling pathway. Collectively, our data provide evidence for further untangling the crosstalk between hypoxia and angiogenesis in cancer.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ANOVA

Analysis of variance

CRC

Colorectal cancer

HIA

Hypoxia induced angiogenesis

ChIP-seq

Chromatin immunoprecipitation sequencing

ssGSEA

Single sample gene set enrichment analysis

SRGA

Signature-related gene analysis

RT-qPCR

Quantitative real-time PCR

Co-IP

Co-immunoprecipitation

MPI

Mannose phosphate isomerase

HIFs

Hypoxia-inducible factors

LDHA

Lactate dehydrogenase A

JAK2/STAT3

Janus kinase 2/signal transducers and activators of transcription 3

CMS

Consensus molecular subtypes

TME

Tumor microenvironment

VEGF

Vascular endothelial growth factor

PDGF

Platelet-derived growth factor

SDF1

Stromal cell-derived factor-1

PGF

Placental growth factor

FGF

Fibroblast growth factor

ANGPT2

Angiopoietin 2

MMP

Matrix metalloproteinase

MDSCs

Myeloid-derived suppressor cells

CAFs

Cancer-associated fibroblasts

TPM

Transcript per million

scRNA-seq

Single cell RNA-sequencing

DEG

Differentially expressed gene

MsigDB

Molecular Signatures Database

KEGG

Kyoto Encyclopedia of Genes and Genomes

GO

Gene Ontology

TSS

Transcription start site

FDR

False discover rate

TFs

Transcription factors

CDF

Cumulative distribution function

PCA

Principal component analysis

MMR

Mismatch repair

UMI

Unique molecular identifiers

UMAP

Uniform Manifold Approximation and Projection

HUVECs

Human umbilical vein endothelial cells

shRNA

Short hairpin RNA

PBS

Phosphate-buffered saline

PVDF

Polyvinylidene difluoride

IHC

Immunohistochemistry

TISIDB

Tumor-immune system interactions database

Treg

Regulatory T cell

pDC

Plasmacytoid dendritic cell

PPI

Protein-protein interaction

Author contributions

Conceptualization: S Liu, Y Zhang, and J Han. Data curation, analysis, interpretation, and visualization: S Liu and Y Zhang.Performing the experiments: S Liu, Y Zhang, Y Meng, and Q Huang. Administrative, technical, or material support: Z Feng, L Wen, X Yang, YG Zhang, and L Qiu.Writing and review of the manuscript: S Liu, Y Zhang, Y Meng, and J Han. Funding acquisition: S Liu, Y Meng, and J Han. Supervision: Z Wang, B Zhang, Z Chen, and J Han. Final approval of the manuscript: All authors.

Funding

This work was supported by the Foundation for Innovative Research Groups of the National Natural Science Foundation of China (81820012), the Natural Science Foundation of Sichuan Province (2023NSFSC0719), Sichuan Science and Technology Program (MZGC20240001), the Fellowship of China National Postdoctoral Program for Innovative Talents, China Postdoctoral Science Foundation (BX20230244), and 1·3·5 Project for Disciplines of Excellence, West China Hospital (ZYGD23017), Sichuan University.

Data availability

All data generated or analyzed in this study are publicly available and can be acquired from UCSC Xena (https://xenabrowser.net/datapages/), GDC (https://gdc.cancer.gov), and GEO (https://www.ncbi.nlm.nih.gov/geo/) with accession number: GSE37892, GSE35896, GSE161158, GSE14333, GSE39582, GSE143985, GSE119409, GSE13294, GSE170999, GSE17356, GSE17357, GSE158632, GSE200204, GSE89831, GSE200203, and GSE205506. Spatial transcriptome was obtained from an open-resource platform (10.5281/zenodo.7551712). SRGA is available on the Github website (https://github.com/LucasLiu20200131/SRGA/). Any additional information is available upon reasonable request.

Declarations

Ethics approval and consent to participate

All animal experimental procedures were in accordance with protocols approved by the Experimental Animal Ethics Committee of West China hospital, Sichuan University (Ethics record number 20220107010).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.